Inferred Network: multiple edges between same source-target pair

[gianfilippo]

Hi,

following the DEA for CCC tutorial, I am having difficulty in properly saving the resulting network
The call
cn.methods.carnival.visualize_network(df_res)

result in a graphviz.graphs.Digraph object

I save it to a file doing the following
p = cn.methods.carnival.visualize_network(df_res)
p.render(locplotDIR + '/potential_regulatory_cascade', format='pdf')

As a result I get 2 files, one is the PDF (attached). You can see I get multiple edges for the same pair of nodes, some edges are black, some blue and some red.

I have multiple edges because the resulting df_res , from running li.mt.find_causalnet, has multiple entries for the same source-target pair. For instance, the Insr-Ptpn11 has 5 edges, 4 red and one black. The df_res shows
source source_type source_weight source_pred_val target target_type target_weight target_pred_val edge_type edge_pred_val
33 Insr input 1.08968 1.0 Ptpn11 unmeasured 0.0 1.0 1 1.0
31 Insr input 1.08968 1.0 Ptpn11 unmeasured 0.0 1.0 1 1.0
29 Insr input 1.08968 1.0 Ptpn11 unmeasured 0.0 1.0 1 1.0
30 Insr input 1.08968 1.0 Ptpn11 unmeasured 0.0 1.0 1 1.0
28 Insr input 1.08968 1.0 Ptpn11 unmeasured 0.0 1.0 1 1.0

I should also point out that I had to run li.mt.find_causalnet using the SCIPY solver, for a license version issue on the cluster.
df_res, problem = li.mt.find_causalnet(prior_graph, input_scores, output_scores, node_weights, node_cutoff=0.1, max_penalty=1, min_penalty=0.01, edge_penalty=0.1, verbose=True, max_runs=50, stable_runs=10, solver='SCIPY')
The solutions are described as Optimal.

I am not sure why I get the same row multiple times. And I am not sure about the color scheme. Could you please clarify ?

Also, is 'render' the best way to save the network to file as image ?

Thanks

[dbdimitrov]

Hi @gianfilippo,

I really suggest using the gurobi solver - it's likely that spicy's inbuilt one (HIGHs) is just resulting in very unstable results. @pablormier can maybe confirm or suggest how to improve the solutions by HIGHs as I haven't used it much (besides running the tests and for the simple example in the tutorial).

I explain how to obtain access to gurobi, here:
https://liana-py.readthedocs.io/en/latest/notebooks/targeted.html#Installing-the-Gurobi-Solver:-A-Step-by-Step-Guide

Hope this helps!

[gianfilippo]

The system manager updated the Gurobi license file. Now things look ok.

In general, I have to say that I am not sure I understand the choice to use the stat without considering the p-value. I feel that the interèpretation of the signalling pathways becomes more difficult.